Author Affiliations
1School of Internet of Things Engineering, Jiangnan University, Wuxi, Jiangsu 214122, China2Faculty of Computer & Software Engineering, Huaiyin Institute of Technology, Huaian, Jiangsu 223003, China;show less
Fig. 1. Detection diagram of R-FCN
Fig. 2. Separable convolution
Fig. 3. Incomplete information marked by anchors
Fig. 4. Change of translation and scaling
Fig. 5. Comparison of iterations accuracy changes
[11]. (a)FPN iterations accuracy changes; (b)R-CNN iterations accuracy changes
Fig. 6. Comparison of pooling. (a) ROI pooling; (b) precise ROI pooling
Fig. 7. Non-maximum suppression
Fig. 8. Lack of localization confidence of NMS
Fig. 9. Prediction of IOU
Fig. 10. Flow chart of LOF-FCN
Fig. 11. Comparison of fire missed detection rate
Fig. 12. Comparison of fire detection accuracy
Fig. 13. Experimental video. (a) Video 1; (b) video 2; (c) video 3; (d) video 4; (e) video 5; (f) video 6
Method | Kernel | Speed /(frame/s) | Accuracy /% |
---|
R-FCN | 13×13 | 15 | 96.9 | Separableconvolution | 13×1,1×13 | 22 | 96.4 |
|
Table 1. Comparison of the speed of the detection
Category | Picture | Number |
---|
L-Fire | 1223 | 2104 | S-Fire | 3512 | 5921 | Total | 4735 | 8025 |
|
Table 2. Training data
Model | YOLOv3 | Fast-RCNN | Faster-RCNN | R-FCN | LOF-FCN |
---|
Speed /(frame/s) | 34 | 0.6 | 7 | 13 | 17 | mAP /% | 76.2 | 54.0 | 83.9 | 80.3 | 81.7 |
|
Table 3. Comparison of algorithm performance
Video | Video frame | Fire frame | Method in Ref. [6] | Method in Ref. [8] | Method in Ref. [9] | Proposed method |
---|
TP /% | FP /% | TP /% | FP /% | TP /% | FP /% | TP /% | FP /% |
---|
1 | 170 | 156 | 96.2 | 3.8 | 97.6 | 2.4 | 98.6 | 1.4 | 99.2 | 0.8 | 2 | 234 | 257 | 94.5 | 5.5 | 93.2 | 6.8 | 92.7 | 7.3 | 95.4 | 4.6 | 3 | 200 | 182 | 99.3 | 0.7 | 99.8 | 0.2 | 100 | 0 | 100 | 0 | 4 | 124 | 93 | 99.7 | 0.3 | 100 | 0 | 100 | 0 | 99.0 | 1.0 | Average | 182.0 | 172.0 | 97.4 | 2.6 | 97.7 | 2.3 | 97.8 | 2.2 | 98.4 | 1.6 |
|
Table 4. Detection results of flame
Video | Video frame | Fire frame | Method in Ref. [6] | Method in Ref. [8] | Method in Ref. [9] | Proposed method |
---|
TN /% | FN /% | TN /% | FN /% | TN /% | FN /% | TN /% | FN /% |
---|
5 | 420 | 0 | 98.8 | 1.2 | 97.3 | 2.7 | 95.4 | 4.6 | 98.3 | 1.7 | 6 | 376 | 0 | 97.2 | 2.8 | 96.2 | 3.8 | 96.7 | 3.3 | 97.6 | 2.4 | Average | 398.0 | 0.0 | 98.0 | 2.0 | 96.8 | 3.2 | 96.1 | 3.9 | 98.0 | 2.0 |
|
Table 5. Detection results of non-flame